MGrounding-630k
收藏资源简介:
MGrounding-630k数据集是由北京交通大学、华中科技大学和清华大学联合创建的大规模多图像接地任务数据集。该数据集包含63万条数据,涵盖了多种多图像接地任务,数据来源于现有数据集和新生成的自由形式接地指令数据。数据集创建过程包括从现有数据集中提取任务数据,并生成新的指令数据以增强模型的接地能力。该数据集主要用于训练和评估多模态大语言模型在多图像接地任务中的表现,旨在解决复杂多图像场景中的精确接地问题,应用于自动驾驶、监控系统和机器人目标定位等领域。
The MGrounding-630k dataset is a large-scale multi-image grounding task dataset jointly created by Beijing Jiaotong University, Huazhong University of Science and Technology, and Tsinghua University. It contains 630,000 data entries covering a variety of multi-image grounding tasks, with data sourced from both existing datasets and newly generated free-form grounding instruction data. The dataset construction process includes extracting task data from existing datasets and generating new instruction data to enhance the grounding capability of models. This dataset is primarily used to train and evaluate the performance of multimodal large language models in multi-image grounding tasks, aiming to solve the problem of precise grounding in complex multi-image scenarios, and has applications in fields such as autonomous driving, surveillance systems, and robotic target localization.




